Beautified frontpage plots and two pylab examples

[vollbier]

This addresses bug #7704: The reason for the thick lines in the frontpage plots seem to have to do with figure size. I left the default figsize ([6.4, 4.8]) and changed the dpi parameter in savefig() from 100 to 25 dpi (which is results roughly in the same number of pixels in the image).

I move the generated images by hand to the doc/_static directory. Should this happen automatically somehow?

I also tried to make the polar_demo.py and pie_demo_features.pylook nicer.
It does not seem possible to pass a list of strings to autopct parameter of pie(). Thus the hack to change the text properties afterwards. I'm not sure, if that's suited for a showcase. I also could delete it, as @efiring suggested. Furthermore, I'm not completely sure, if the radial positioning of the labels in pie() is always correct.

I did not succeed at changing the text appearance in table_demo.py.

[tacaswell]

Can you either match the pixel size exactly or change the values in _static/mpl.css div.responsive_screenshots to match these sizes?

It would be better if it automatically copied and regenerated these on every build.

[codecov-io]

Current coverage is 62.12% (diff: 100%)

Merging #7740 into master will not change coverage

@@             master      #7740   diff @@
==========================================
  Files           174        174          
  Lines         56028      56028          
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
  Hits          34805      34805          
  Misses        21223      21223          
  Partials          0          0          

Powered by Codecov. Last update 1fa4dd7...cc66626

[tacaswell]

@vollbier Could you squash this into a single commit (to avoid having the extra binary images in the history)? I can also do it if you like.

[vollbier]

@tacaswell No problemo.

[phobson]

I approved this as well, but would like another set of eyes on it. I was able to build the docs and things indeed look better, but this touches enough sphinx code that I hesitate to merge it myself.

[NelleV]

I really recommand against change the make.py. I think the goal is to move to the "modern" way for sphinx to build, and in additition of making our current documentation build system hard to maintain and not standard, this will break when we finalize the migration to make.

[NelleV]

Understand my last comment as: I am strongly against part of this patch unless it builds upon our current gallery system and avoids modifying the make.py.

[tacaswell]

I am not super concerned about the changes to make.py.

make.py is a big chunk of technical debt no matter what way you look at it. It we want to auto-generate the main page images (which are very different than a gallery example) we will have to have a custom build target or write a sphinx extension to do it. When we move to the modern makefile we can decide to keep this complexity (because auto-generating the main page images is a pretty reasonable thing to do) or drop it and go back to static images. In either case, there is little future cost to this now.

[vollbier]

I don't have any strong opinion about modifying make.py or not. I took a look at the current doc building structure: It seems like a big chunk of work to get it "sphinx-conform" - it's probably better to do that in a separate pull request.

[NelleV]

@tacaswell I disagree with you. Most scientific projects manage just fine with normal sphinx extensions and sphinx.

Is there an interest in moving to the up-to-date build system?

[dopplershift]

I think the discussion of what's needed for matplotlib's docs going forward is easily derailed in this forum, as well as off-topic for this PR. I vote we punt that to one of the weekly dev hangouts (open to anyone interested).

As far as this PR is concerned, I agree make.py is a mess, so let's just get the changes in it necessary to get the front page how we want for 2.0.

[NelleV]

There is absolutely no need to change the make.py for the sake of this PR. The images can be generated using the gallery system, and linked back properly to the front page if necessary. The reason I did not do this in my original PR updating those plots is that this will add 4 more plots to the gallery, which already has hundreds of example and it seemed like not such a good idea.

[tacaswell]

I suggested in my second comment that it would be better to automatically generate these files (which @vollbier did in a very direct way). As I said above, I do not think this change moves the needle on the level of technical debt we have in make.py and is an improvement to building the docs.

make.py is a bit of an anachronism, but it works reliably. Updating it is probably a good idea, but not near the top of my internal list of priorities.

I am strongly in favor of merging this as-is, but given @NelleV 's concerns would at least one more senior dev to weigh in.

[WeatherGod]

If the changes here works and makes things more resilient to human error (forgetting to generate images, etc.), then I am fine with small changes to make.py to make that happen.

…

On Fri, Jan 13, 2017 at 10:44 AM, Thomas A Caswell ***@***.*** > wrote: I suggested in my second comment that it would be better to automatically generate these files (which @vollbier <https://github.com/vollbier> did in a very direct way). As I said above, I do not think this change moves the needle on the level of technical debt we have in make.py and is an improvement to building the docs. make.py is a bit of an anachronism, but it works reliably. Updating it is probably a good idea, but not near the top of my internal list of priorities. I am strongly in favor of merging this as-is, but given @NelleV <https://github.com/NelleV> 's concerns would at least one more senior dev to weigh in. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#7740 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARy-HMls32QVG18FYB3lvmagHRnSnBqks5rR5vYgaJpZM4LasVf> .

[NelleV]

Changing the make.py is a terrible idea because it is a file that inherently does not belong to matplotlib. The modifications embedded in it has made my life hell in moving forward to the new makefile system, and it is because we keep doing these hacks that the project is so hard to maintain.

There is a way to generate the images without modifying the make.py. I am suggesting using the tool that has been build for this purpose to do this instead of modifying the make.py.

[NelleV]

I really would have appreciated not merging this PR until we reached a consensus.
This breaks my patch for no good reason, and I am not putting more effort into the migration of the makefile.

[efiring]

The real problem with this is not the change to make.py, which is irrelevant, it is that the pie chart demo is still an embarrassment, though not as bad as it used to be.
Here is v2.x:

And below is master after the merge:

Not only is it still "chart junk", it is still incompetent chart junk. The incompetence is in the Matplotlib implementation, not in the writing of the example script. Can we please simply delete this example? Would we want anyone to make a plot like this and advertise it as "made using Matplotlib"?

[tacaswell]

👍 to removing the pie chart demo.

[efiring]

On 2017/01/13 9:19 AM, Nelle Varoquaux wrote: This breaks my patch for no good reason, and I am not putting more effort into the migration of the makefile.

I'm sorry, @Nellev, but which patch are you referring to?

[NelleV]

The migration to the Makefile. I closed the pull request, as the migration to Makefile depends on the migration of sphinx-gallery. The branch is still on my fork if someone wants to pick up the work, but considering the amount of work it requires, I doubt you'll find volunteers.

[phobson]

@NelleV is that #6663?

[NelleV]

Yes. I think the last remaining point is the sphinx-gallery migration. There are only around 350 examples to tackle there.

[efiring]

Maybe the approach we need to take here is to think of the migration as gradually creating a sphinx gallery in parallel, adding new or ported examples as needed to showcase features and capabilities. In other words, eventually abandon the historical legacy of existing examples; what are good examples for Matplotlib now and in the future?

[QuLogic]

Backported to v2.x as 2f3041e.